JP2988097B2 - How to segment a grayscale image - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a region dividing method for dividing a gray image having a plurality of textures (patterns) in an image by an image processing apparatus. In the drawings, the same reference numerals indicate the same or corresponding parts. .

[0002]

2. Description of the Related Art FIGS. 5 and 6 are explanatory diagrams of a conventional method for dividing a gray-scale image of this type into regions. In FIG. 5, (a) is an original grayscale image, (b) is a reduced image of the original image (a), 8 is a texture area # 1, and 9 is a texture area #
2, Z (Z1, Z2) is a rectangular small area, and 11 is a sampling pixel. 6A shows the density histogram of the rectangular small area Z1 in FIG. 5B (that is, the distribution of the number of pixels (vertical axis) for each density (horizontal axis)), and FIG.
FIG. 5B shows a density histogram of the rectangular small area Z2 in FIG.

[0003] Conventionally, as a method of dividing a gray-scale image into regions, a reduced image (b) is created by sampling an original gray-scale image as shown in FIG. Is divided into rectangular small areas Z (Z1, Z2,...) Each side of which has a predetermined number of pixels (three in this example).
The density histogram of each of the small areas Z1 and Z2 is obtained as follows. Next, the density histogram of each small area Z is compared with the density histogram of the adjacent small area Z. If the difference between the two density histograms is equal to or smaller than the reference value, the two small areas Z are regarded as one area. Integrate. This integration process is performed for all the small areas Z in the reduced image to perform area division.

[0004]

However, in the above-mentioned conventional method for dividing a gray-scale image into regions, the texture region as shown in FIG. 5 at 8 is lost due to the value of the sampling interval m when a reduced image is created. There was a problem that there was. Accordingly, it is an object of the present invention to provide a method for dividing a gray-scale image, which can solve such a problem.

[0005]

According to a first aspect of the present invention, there is provided an area dividing method comprising the steps of: converting an original grayscale image in which multivalued pixels are arranged at equal intervals in two directions, horizontal and vertical. Each side is divided into a first rectangular area (Amn or the like) composed of m (x or the like) pixels, and a reduced image composed of one pixel each representing the first rectangular area is divided into n (s) Second rectangular area (eg, rectangular small area Z) composed of pixels
And a density distribution between adjacent ones of the second rectangular area is compared, and when the difference is equal to or smaller than a predetermined value, the two second rectangular areas are integrated. And the density value of one pixel representing the first rectangular area is defined as an average density value (Dmn) of the pixels of the first rectangular area.

A second aspect of the present invention is a method of dividing an area.
In the area dividing method described in (1), a cumulative density histogram (H (d)) for the second rectangular area is used as the amount indicating the density distribution.

[0007]

The original gray image is divided into (x pixels) × (x pixels) rectangular areas Amn, and the average pixel density Dmn in each rectangular area is obtained.
, And a reduced image is created by using a pixel whose average density value is a pixel density value as a pixel representing the corresponding rectangular area Amn. Then, the reduced image is further divided into (s pixels) × (s pixels) rectangular small areas Z, and an area division is performed by comparing the cumulative density histograms H (d) between adjacent small areas Z. As described above, when the reduced image is created, the rectangular area A
The density average value Dmn in mn is obtained, and this value is used as the density value of the pixel of the reduced image, so that the density value of each pixel of the original grayscale image before reduction is reflected. Therefore, even in the reduced image, rough area division can be performed without completely losing the texture information of the image before the reduction.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to FIGS. FIG. 1 is a block diagram showing an apparatus configuration as an embodiment of the present invention. In the figure, 1 is an image input device, 2 is an original image storage unit, 3 is a density calculation unit, 4 is an image reduction unit, 5 is a reduced image storage unit, 6 is a cumulative density histogram creation comparison unit, and 7 is a result storage unit. is there. Next, FIG. 2 to FIG.
FIG. 3 is a diagram for explaining the operation of FIG. 1. Here, FIG. 2 is an image diagram corresponding to FIG. 5, FIG. 2 (a) is an original grayscale image, and 8 and 9 are texture regions # 1 and # 2 similar to FIG. 5 (a), respectively. . FIG. 2 (b) shows the above (a)
Is a reduced image of the original image of Z (Z1, Z2, Z3,
..) Are rectangular small areas in this reduced image. Next, FIG. 3 is an explanatory diagram of a cumulative density histogram described later for the rectangular small areas Z1 and Z3 of FIG. 2B and a density distribution shape verification amount for comparing the histograms, and FIG. 4 is a diagram of FIG. FIG. 14 is a diagram illustrating a region division result of a reduced image.

Next, the operation of each part of FIG. 1 will be described with reference to FIGS. That is, the image input device 1 shown in FIG.
It is assumed that an original grayscale image as shown in FIG. The grayscale image is stored in the original image storage unit 2 and sent to the density average value calculation unit 3 at the same time. In the density average value calculation unit 3,
A rectangular area consisting of a preset size, that is, a pixel number x (three in this example) of each side is defined as one rectangular area Amn (where m and n indicate the coordinates of the area), and a plurality of original images are obtained. The image is divided into rectangular areas, and the average density value Dmn in each area is obtained. That _{Dmn = (d ∈ Amn Σd)} / x 2 where, d: the pixel density value, _(d ∈ _Amn Σd): the sum of the density value d of each pixel of the rectangular region Amn. The density average value Dmn obtained in this way is sent to the image reduction unit 4, where the density average value Dmn of one rectangular area Amn is set as the density value of one pixel representing the rectangular area Amn, as shown in FIG. A reduced image as shown in b) is created.

Next, region reduction is performed on the reduced image of FIG. 2B by using, for example, a method called a statistical hypothesis test method. Hereinafter, a method of segmenting a region based on the statistical hypothesis test method will be described. First, the reduced image of FIG. 2B is divided into rectangular small areas Z (Z1, Z2, Z3,...) Each having the number of pixels s × s (s = 3 in this example) on each side. Next, the density distributions of the adjacent rectangular small areas Z are compared by a method described later, and if the shapes of the density distributions are similar, the two small areas Z
Are integrated into one region. By repeating this integration process until the regions are no longer integrated, a final region division result such as the divided regions 10 and 11 in FIG. 4 is obtained.

Here, for example, a method of comparing the density distribution shapes of two adjacent rectangular small areas Z1 and Z3 in FIG. 2B will be described. Now, let h ₁ (d) and h ₃ (d) be
1, and a density histogram of Z3. From these histograms, cumulative density histograms H ₁ (d) and H ₃ (d) are respectively obtained.
Ask for. However here the general concentration histograms of small rectangular area Z and h (d), the cumulative density histogram H of the region Z (d) is _{H (d) = i = 0} Σ d h
(I).

With respect to the cumulative density histograms H ₁ (d) and H ₃ (d) obtained in this way, for example, the following (a),
(B) concentration distribution shape test amount X by any of the test methods
Xa and Xb are obtained. (A) Kolmogorov-Smirnov test method: X = Xa = _d max | H ₁ (d) -H ₃ (d) | (1) That is, in this case, the test amount Xa is the pixel density the difference H ₁ of the cumulative density histogram when the value d as a variable (d) -H ₃ (d)
Is given as the maximum of the absolute value of. (B) Smoothed Difference _{assay: X = Xb = d Σ |} H 1 (d) -H 3 (d) | ........................... (2) That is, in this case, test weight Xb cumulative density histogram Of the differences H ₁ (d) −H ₃ (d) for all the pixel density values d as variables (that is, equivalent to the area of the shaded portion (total number of pixels) in FIG. 3 described below). Value). Thus, if the density distribution shape test amount X (Xa or Xb) according to the above formula (1) or (2) is a predetermined threshold value Th, if X ≦ Th, the two The rectangular small areas Z1 and Z3 are integrated into one area.

FIG. 3 shows the cumulative density histograms H ₁ (d) and H ₃ (d) obtained as described above, and equations (1) and (2).
An example of the test amount X (Xa, Xb) according to (2) is shown. In this case, the regions Z1 and Z3 can be interpreted as different regions. Then, as shown in FIG. 4, the area distribution results in the reduced image of FIG. 2B are two areas of 10 and 11 in the range shown in FIG. 2B.

The length s of one side of the rectangular small area Z is 5 which is the minimum size at which the area is to be divided from other texture areas.
It is appropriate that the length is about 0%. The case shown in FIG. 2B corresponds to a case where the size of the texture area # 2 is considered to be the minimum size to be divided.

[0015]

According to the present invention, by using a reduced image when performing area division, the amount of calculation for density histogram calculation and rectangular area integration is reduced. Since the density average value obtained from each of the image values is used, rough area division can be performed in the reduced image without completely losing the texture information of the original image before the reduction.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a hardware configuration as an embodiment of the present invention.

FIG. 2 is a diagram showing an example of an original image and its reduced image for explaining the operation of FIG. 1;

FIG. 3 is a diagram showing an example of a cumulative density histogram and a density distribution shape test amount for explaining the operation of FIG. 1;

FIG. 4 is a diagram showing an example of an area division result based on FIG. 1;

FIG. 5 is a diagram showing a conventional original image corresponding to FIG. 2 and a reduced image thereof.

FIG. 6 is a diagram showing a density histogram of a rectangular small area in FIG. 5;

[Explanation of symbols]

Reference Signs List 1 Image input device 2 Original image storage unit 3 Density average value calculation unit 4 Image reduction unit 5 Reduced image storage unit 6 Cumulative density histogram creation / comparison unit 7 Result storage unit 8 Texture area # 1 9 Texture area # 2 10 Division area 11 divided region Z (Z1, Z2, Z3, ...) small rectangular area Amn rectangular area Dmn average density value _{H (d) (H 1 (} d), H 3 (d) cumulative density histogram X (Xa, Xb) concentration distribution shape Test amount

Claims (2)

(57) [Claims] An original grayscale image in which multi-valued pixels are arranged at equal intervals in two directions, horizontal and vertical, is divided into a first rectangular area having m pixels on each side. The reduced image composed of one pixel each representing a rectangular area is divided into a second rectangular area having n pixels on each side, and the density distributions of adjacent ones of the second rectangular area are compared with each other. When the difference is equal to or less than a predetermined value, the method is an area dividing method of a gray image in which the two second rectangular areas are integrated, and the density value of one pixel representing the first rectangular area is determined by the first first area. An area dividing method for a gray-scale image, characterized in that the average value is the average density of pixels in a rectangular area. 2. The area dividing method according to claim 1, wherein
A method for dividing a gray-scale image area, comprising using a cumulative density histogram for the second rectangular area as the amount indicating the density distribution.